✦ PromptlyBase.com
SK EN
✦ The Prompting Guide

How to Prompt AI
Like a Pro

Learn to communicate with Claude, ChatGPT and Gemini so they deliver exactly what you need — on the first try.

⏱ 25 min read 🧪 5 real-world examples 🎯 Beginner to advanced
0

What is a Prompt?

Theory

A prompt is the message you send to an AI model. It can be a question, a command, a task description, or any combination. The quality of your prompt directly determines the quality of the response — the golden rule applies:

Garbage in, garbage out. The more precisely you describe what you want, the better the result. AI is not a mind reader — it's an extremely capable assistant that needs context to do its best work.

Prompting is a skill you can learn and improve. People who prompt well extract 10× more value from AI than those who just type "write me an email."

1

How AI "Thinks"

Theory

Modern AI models (Claude, ChatGPT, Gemini) are Large Language Models (LLMs). They weren't programmed with rules — they learned from vast amounts of text to predict which word (token) most likely comes next.

This has real implications for prompting:

  • Context is king. More relevant context leads to better answers.
  • Format propagates. If you start with bullet points, AI continues with bullet points.
  • AI can't read minds. What you don't mention, it doesn't know.
  • Role changes output. "You are an expert in..." shifts the tone and depth significantly.
💡 Think of it this way: AI is like a brilliant new employee on day one. They know a lot, but don't know your context, your clients, or your preferences — until you tell them.
2

Anatomy of a Great Prompt

Theory

A great prompt typically has these components — not all are always needed, but the more you include, the better the output:

1 Role (Persona)

Tell AI who to be. This sets the tone, depth, and vocabulary of the response.

"You are a senior tax advisor specializing in small businesses..."
2 Context

Describe the situation — what's happening, who you are, what the goal is.

"I run an e-commerce store with 500 monthly orders. Customers complain about long delivery times..."
3 Task

Tell AI exactly what to do. Use action verbs: write, analyze, compare, suggest, translate, summarize.

"Suggest 3 specific solutions to reduce delivery time below 3 days."
4 Output Format

Specify how the answer should look — list, table, JSON, email, word count, etc.

"Answer in bullet points, max 2 sentences each, no intro paragraph."
5 Constraints / What to Avoid

Tell AI what to leave out — saves time on corrections.

"No jargon, skip the intro, exclude solutions over $10,000."
Full prompt example:
"You are an experienced logistics consultant. I run a clothing e-commerce store with 500 monthly orders and customers complain about 5–7 day delivery. Suggest 3 concrete solutions to get it under 3 days. Answer in bullet points, max 2 sentences each, no intro, exclude solutions over $10,000."
3

Prompting Techniques

Theory

Zero-shot prompting

Give a direct command without examples. Works well for simple tasks.

Translate this to French: "Good morning, I would like to book a table."

Few-shot prompting

Show 2–3 input/output examples. AI recognizes the pattern and applies it.

Classify reviews as POSITIVE / NEGATIVE / NEUTRAL.

Review: "Great product, arrived on time!" → POSITIVE
Review: "Packaging was damaged." → NEGATIVE
Review: "Product is okay." → NEUTRAL

Review: "Quality is good but delivery took a week." →

Chain-of-thought (CoT)

Ask AI to "think out loud." Improves results for logic and math problems.

Solve this step by step:
If I sell 150 units at $12 each and total costs are $1,200,
what is my profit margin as a percentage?

Role prompting

Assign AI a specific role, expertise, or personality.

You are a senior UX designer with 10 years in SaaS products.
Critique my onboarding flow from a user perspective:
[paste your flow description]
4

Claude vs ChatGPT vs Gemini

Theory

Each model has different strengths. A prompt that works great on one won't always give the same result on another.

Feature Claude (Anthropic) ChatGPT (OpenAI) Gemini (Google)
Long documents ✓ Excellent ◐ Good ◐ Good
Coding ✓ Excellent ✓ Excellent ◐ Good
Real-time info ✗ Limited ✓ Web search ✓ Google integrated
Safety / Ethics ✓ Strictest ◐ Moderate ◐ Moderate
Multimodal (images) ✓ Yes ✓ Yes ✓ Yes
Best for Analysis, writing, code General purpose Google Workspace integration
💡 Practical recommendation: Start with ChatGPT or Claude. For long documents and deep analysis, try Claude. For questions that need current data, try Gemini or ChatGPT with web search enabled.
5

Example: Business

Example

Scenario: Writing a polite rejection email to a client asking for a discount

You need to decline gracefully, preserve the relationship, and avoid sounding like a template.

❌ Weak prompt
Write an email declining a discount request.
✅ Strong prompt
You are an account manager at a B2B software company.
Client (John Smith, TechLogix Inc.) is requesting a 20% discount
on an annual license ($4,800/year). This isn't possible,
but we want to retain the client.

Write a rejection email that:
- Feels personal, not templated
- Explains why a discount isn't possible (no excuses)
- Offers an alternative (payment plan or bonus features)
- Preserves a positive relationship
- Is max 150 words, formal but friendly tone
Why is this better? The strong prompt gives context (B2B, contract value, client name), defines what the email must contain, and sets the format. AI doesn't guess — it executes your exact specification.
6

Example: Healthcare

Example

Scenario: A doctor needs to explain a diagnosis in plain language

Patient had a hypertensive crisis — the doctor wants a clear, jargon-free discharge explanation.

✅ Strong prompt
You are a doctor explaining a diagnosis to a patient with no medical background.

Diagnosis: Hypertensive crisis (BP 195/115 mmHg on arrival)
Patient: 58-year-old male, teacher, no previous treatment

Write a patient explanation that:
1. Explains what happened and why it's serious (no scaremongering)
2. Describes how treatment will proceed
3. Lists 3 lifestyle changes that will help
4. Answers "can I still drive?"

Use plain language (no Latin terms), max 200 words, divided into short paragraphs.
⚠️ Important disclaimer: AI is not a replacement for a doctor. AI outputs in healthcare must always be reviewed by a qualified professional. Use AI to assist in preparing materials, not as a diagnostic source.
7

Example: Education

Example

Scenario: A teacher creates a history quiz for 8th grade

They want a mix of question types — not just fact memorization.

✅ Strong prompt
You are an experienced middle school educator.

Create a history quiz on "World War II" for 8th grade students.

Requirements:
- 10 questions total
- Mix: 4× multiple choice (A/B/C/D), 3× short answer, 2× true/false, 1× essay (5–7 sentences)
- Questions test understanding of causes and consequences, not just dates
- Difficulty: 3 easy, 5 medium, 2 hard
- Include an answer key at the end
- Age-appropriate vocabulary
💡 Bonus tip: Add "Explain why each answer is correct" — AI will generate a teacher's commentary you can use in class discussions.
8

Example: Creative Writing

Example

Scenario: A marketer creates an Instagram series for a specialty coffee shop

They want a consistent brand voice, not generic captions.

✅ Strong prompt
You are a social media copywriter specializing in F&B brands.

Coffee shop: Grind & Co., Brooklyn NY
Target audience: 25–40, creatives, remote workers
Brand voice: friendly, slightly poetic, never corporate
Emphasis on: local sourcing, slow coffee, the story behind each cup

Create 5 Instagram posts for a work week (Mon–Fri):
- Each post: caption (max 120 chars) + 5 hashtags
- Monday: motivational, Tuesday: educational (about coffee),
  Wednesday: product spotlight, Thursday: behind-the-scenes,
  Friday: weekend mood
- Never use clichés like "Good morning ☀️" or "We love coffee ❤️"
9

Example: Coding

Example

Scenario: A developer needs a code review on their function

They don't just want "fix bugs" — they want structured, constructive feedback.

✅ Strong prompt
You are a senior Python developer focused on clean code and security.

Do a code review of this function. Stack: Python 3.11, FastAPI, PostgreSQL.
Context: This is an API endpoint for registering new users.

```python
async def register_user(email: str, password: str, db: Session):
    user = db.query(User).filter(User.email == email).first()
    if user:
        return {"error": "exists"}
    new_user = User(email=email, password=password)
    db.add(new_user)
    db.commit()
    return {"ok": True}
```

Focus on:
1. Security vulnerabilities (top priority)
2. Missing validations
3. Error handling
4. Performance

For each issue: describe what's wrong, why it's a problem, show the fix.
⚠️ Spot the bug? The password is stored as plaintext — a critical security vulnerability. A good code review prompt will always catch this.
10

Common Mistakes

Tips
Too vague

"Help me with marketing" → AI doesn't know what you want. Always specify: what, for whom, in what format.

Missing context

"Write an email" → To whom? About what? What tone? Without context you get a generic output.

Too many tasks at once

One prompt = one task. If you want an analysis AND solutions AND a follow-up email — split into 3 prompts.

Accepting outputs uncritically

AI makes mistakes. Always verify facts, numbers, and citations. AI is an assistant, not an oracle.

Giving up after one attempt

Prompting is a conversation. If the first answer isn't perfect, follow up: "Focus more on X", "Cut it in half", "Add examples".

11

Cheat Sheet — Prompt Template

Tips

Copy, fill in, and send:

You are [role / expert in X].

Context: [describe the situation — who you are, what's happening, what the goal is]

Task: [what exactly AI should do — write, analyze, compare, suggest, translate...]

Input/material: [paste the text, data, or code for AI to work with]

Output format: [list / table / email / JSON / word count...]

What I don't want: [what to leave out, what to avoid]

Want to go deeper?

Once you've mastered prompting, the next step is AI agents — autonomous systems that use tools and make decisions.

simplyagents.net →